|
on Artificial Intelligence |
| By: | Po Han Teo |
| Abstract: | Large Language Models (LLMs) are increasingly used as stand-ins in behavioural games. These stand-ins rely on the assumption that the LLM's distribution of choices meaningfully matches how humans play the same game. This study tests that assumption through two games. The first is a p-beauty contest, and the second one is a public goods game. The study first investigates five local-model settings within the same model family. These settings are varied together in a 360-cell factorial, which balances temperature, scale (0.5-32B), quantisation, instruct vs base, and framing. Each cell's distribution is then compared against whole choice distributions in published human data. Each deployment setting, except for quantisation, governs a different aspect of fidelity. Mechanically, while the dispersion of human players can be somewhat recovered through deployment settings, the strategic process behind it cannot. Through the lens of the level-k cognitive theory, we find that LLMs act as static, category-retrieved level-k players, where k is set by the model scale. The models also do not run within-game belief-updating or backward induction throughout multiple-round horizon settings. While human contributions decayed in the public goods game, LLMs stayed flat or rose at every scale. When the horizon test was administered, LLMs were more cooperative under an indefinite horizon compared to a finite one. However, LLMs ignore their relative round position, so no last-round defection was displayed. This implies that LLMs retrieved levels relative to the horizon category rather than working out iteratively from the specific game setting. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.27845 |
| By: | Qiaoni Shi; Kai Zhu; Kai Gu |
| Abstract: | Search engines have long allocated attention on the web by routing users from queries to websites. AI search changes this arrangement because information needs can be resolved inside the intermediary. Using URL-level Comscore U.S. desktop clickstream, we compare ChatGPT and Google information-seeking occasions and exploit ChatGPT Search access expansions to estimate traditional search displacement. ChatGPT produces outbound clicks in only 5.2% of conversation sessions, far below Google's referral ratio. The remaining clicks are not a scaled-down Google stream: they skew toward specialized destinations and away from ad-supported sites. Wider access cuts search use by 9.4%, with search-referral losses largest for informational categories. Our findings identify a central economic shift in digital intermediation: AI search might satisfy information needs inside the intermediary while weakening the referral bargain that has linked search, traffic, and content production on the open web. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.07652 |
| By: | Zara Contractor; Germ\'an Reyes |
| Abstract: | We study how generative AI affects student learning in a randomized experiment. In proctored, in-person sessions, undergraduates learn about an unfamiliar topic and write an analytical essay with or without access to off-the-shelf generative AI, then complete unaided assessments immediately and one week later. We measure learning with knowledge tests (factual and conceptual understanding) and open-ended essays (higher-order skills). AI access raises immediate test scores by 0.27 standard deviations. These gains persist one week later. Essay quality, by contrast, changes little while students have AI access but improves in style and relevance one week later, when students write unaided. These delayed gains are larger among augmentation users-who use AI to explain concepts rather than generate text-whereas automation users' short-run quality gains vanish once AI is removed. We find evidence for two mechanisms behind the learning gains: students shift time away from drafting text and toward reading and searching for information, and they report greater learning enjoyment. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.08849 |
| By: | Nicola Borri; Yukun Liu; Aleh Tsyvinski |
| Abstract: | Using 380 trillion tokens of realized AI consumption across more than four hundred large language models from the licensed proprietary OpenRouter dataset covering approximately 2 percent of current global monthly AI token consumption, we analyze how AI affects firms, markets, and workers. Leveraging the unprecedented size, scope and granularity data, we construct the AI Factor from growth in tokens, dollars, and users, estimate firm-level AI Betas from stock return comovement, and characterize the AI Premium. First, we build a high-frequency AI factor and decompose it into salient components. Second, we show that firms whose returns covary more positively with the AI factor--high AI beta firms--earn higher subsequent returns, and the AI premium is large and heterogeneous. A value-weighted long-short strategy earns 64.1 basis points per week, and the premium is large for loadings on the intensive, frontier-oriented margin of AI consumption-closed-source models, paying and seasoned users, and long prompts--but not on casual or open-weight use. Third, the premium reaches beyond technology firms into consumer-facing and capital-heavy parts of the economy, but is absent in emerging markets, including China. Fourth, the AI exposure is more positive in nonroutine interactive work and the more negative in analytical, scientific, and operations-control skills--an occupation one standard deviation higher in interaction-and-communication content has 0.36-standard-deviation higher market-implied AI premium. Additionally, we provide early evidence of the rise of the agentic economy. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.30583 |
| By: | Auyon Siddiq; Niuniu Zhang |
| Abstract: | Has generative AI changed how labor markets value human capital? We study this question using data from Upwork, a large online labor market. Representing worker profiles with high-dimensional text embeddings, we compute the importance of human capital information and price in predicting labor demand, and incorporate these measures into a difference-in-differences design around the release of ChatGPT. We find that in more AI-exposed job categories, the importance of human capital declines and the importance of price rises, suggesting a commoditization effect of AI on labor. Two additional findings support commoditization as a mechanism: The demand premium enjoyed by workers with strong human capital declines in more AI-exposed categories, and demand reallocates toward lower-priced workers. Our results have implications for the design of online labor markets, workers' incentives to invest in human capital, and labor welfare. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.21880 |
| By: | Simrita Singh; Naireet Ghosh; Tinglong Dai |
| Abstract: | When firms deploy autonomous AI, they must decide how much work to leave to the system and how much to keep workers engaged. This decision affects current output and future human capital. We develop a parsimonious two-period model in which AI may outperform the worker when it functions, but may fail with positive probability. A firm chooses worker engagement; engagement lowers current output for below-benchmark workers, but changes future skill through learning and erosion. We distinguish two dimensions of AI progress: capability, the system's output when it works, and reliability, the probability that it works. In a single-firm benchmark, engagement is valuable only as fallback investment. The firm engages the least-skilled workers most, because they have the largest skill gaps and are least costly to bring toward a useful fallback level. With worker mobility, engagement also affects labor-market sorting: workers prefer jobs that build more valuable skill trajectories. This sorting motive targets higher-skill workers near the AI frontier, where skill gains are more valuable and engagement is less costly. Mobility can therefore reverse the engagement pattern, shifting investment from the least-skilled toward the most-skilled workers below the AI benchmark. Mobility also reshapes how AI progress affects engagement: greater capability raises engagement by increasing the value of the skill trajectory a firm offers, whereas greater reliability can raise or lower it because it reduces fallback need while also changing learning opportunities. Under worker mobility, human-AI work design becomes a problem of human-capital investment, in which allocating work today shapes future skill. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.29111 |
| By: | Yang Yu; Martin Fleming; Lucy Hampton; Christophe Combemale; Neil Thompson |
| Abstract: | The adoption of artificial intelligence (AI) by large enterprises is an important potential source of aggregate productivity improvement and labor market impact. We study AI adoption of S&P 500 firms over the period 2016 to 2025, estimating adoption at the enterprise level. While generative AI tools are useful for personal and professional applications, our focus is on the deep integration of AI in the business processes of large enterprises which are bellwethers for firm adoption more broadly. We develop a novel measure to assess deep AI adoption (and distinguish it from AI hype) that is based on SEC 10-K filings, where laws and regulations ``prohibit companies from making materially false or misleading statements." In 2025, 11% of S&P 500 enterprises had AI deeply integrated into their business processes, and a further 10% were using AI in the production of goods and delivery of services. AI adoption has more than quadrupled from 5% in 2022 with slowly accelerating adoption among non-technology firms but very aggressive adoption in the technology sector which accounts for two-thirds of deeply integrated enterprise adoption. Firm profitability shows a "J-curve" as firms move from no adoption to deep adoption, but we observe no differences in capex or productivity. Among technology firms, but not others, AI adoption is higher for firms with more employees and higher values of Tobin's q. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.08920 |
| By: | Gambacorta, Leonardo; Jappelli, Tullio; Oliviero, Tommaso |
| Abstract: | We present findings from a specialized module on generative artificial intelligence (gen AI) included in the Italian Survey of Consumer Expectations (ISCE), conducted in 2024 with a representative sample of Italian individuals. This analysis offers novel insights into current and anticipated interactions with gen AI tools and the potential benefits from adoption. As of April 2024, 75.6% of the Italian population aged 18–75 was aware of gen AI, 36.7% had used it in the previous 12 months, and 20.1% reported monthly usage. Socio-economic factors significantly influence adoption rates, with higher usage observed among men, individuals with college degrees, and younger individuals, particularly students. Looking ahead, gen AI is expected to be used more frequently for education and leisure activities in the coming months. Finally, using a Mincer earnings regression, we highlight that the income return associated with gen AI usage is around 2%. |
| Keywords: | Generative AI; Household survey |
| JEL: | D10 O33 |
| Date: | 2025–10 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20762 |
| By: | Zhong, Hongda |
| Abstract: | I study the optimal integration of humans and technologies in multi-layered decision-making processes. Each layer can correct existing errors but may also introduce new ones. A one-dimensional quality metric – a decision-maker’s error correction capability normalized by its new errors – determines the optimal rule: deploying higher-quality technologies in later stages. Interestingly, the final decision-making layer may not achieve the greatest error reduction; instead, its role hinges on minimizing new errors. Human effort varies asymmetrically across layers—early stages prioritize error correction with lower effort, while later stages emphasize avoiding new errors with higher effort. Applying the model to artificial intelligence (AI) reveals that AI's generative capabilities make it more likely to serve as the final decision-maker, reducing the need for costly human input, but underscoring the risks of AI hallucination. The theoretical framework also extends to applications including repeated delegation, automation design, loan screening, tenure review, and other multi-layer decision-making scenarios. |
| Keywords: | Automation; Delegation |
| JEL: | C44 M51 |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20330 |
| By: | Feher, Adam; Garcia-Appendini, Emilia; Mihet, Roxana |
| Abstract: | We leverage a novel dataset on U.S. data center energy loads, utility electricity prices, and establishment-level revenues, employment, and carbon emissions from 2010 to 2023 to examine whether rising data center demand affects local retail energy prices or other spillovers. For identification, we employ an instrumental variables continuous difference-in-differences design, exploiting exogenous variation in data center location attractiveness. We find no detectable local spillover effects from data center energy growth. A regional model calibrated to these null results suggests that shocks larger than those observed through 2023 could still result in noticeable increases in household utility bills if not offset by regulation or external supply. |
| Keywords: | Climate change; Technology adoption; Data Centers |
| JEL: | Q55 Q58 O44 L94 |
| Date: | 2025–10 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20758 |
| By: | Bonfiglioli, Alessandra; Crinò, Rosario; Filomena, Mattia; Gancia, Gino |
| Abstract: | We study the environmental impact of artificial intelligence (AI) using a novel dataset that links measures of AI penetration, the location of data centers and power plants, and CO2 emissions across US commuting zones between 2002 and 2022. Our analysis yields four main findings. First, exploiting a shift–share identification strategy, we show that localities more exposed to AI experience relatively faster emissions growth. Second, decomposition results indicate that scale effects dominate, while changes in industrial composition exert at most a weak mitigating effect; at the same time, electricity generation becomes more carbon intensive. Third, AI penetration raises dependence on non-renewable electricity. Fourth, proximity to data centers is a key driver of this effect, as nearby power plants shift toward greater fossil fuel use. These findings suggest that, absent a rapid decarbonization of power generation, the diffusion of AI is likely to exacerbate environmental externalities through the energy demand of data centers. |
| Keywords: | Data Centers; Environment; Emissions; Pollution |
| JEL: | O33 Q55 R11 |
| Date: | 2025–09 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20686 |
| By: | Yiqing Wang; Yixin Kang; Luyun Lin; Siqi Mao |
| Abstract: | The release of SR 26-2 marks a significant modernization of U.S. model risk management by replacing SR 11-7 with a more risk-based and materiality-sensitive supervisory framework. However, generative and agentic AI are excluded, creating an important governance challenge for banking organizations and other financial institutions. Although generative AI may not directly estimate credit risk or make underwriting decisions, its outputs can materially affect the surrounding control environment through monitoring interpretation, policy analysis, or adverse-action language drafting. These uses may influence how regulated financial decisions are explained, challenged, documented, and governed. This paper proposes the Generative AI Control Framework (GAICF), an SR 26-2-compatible governance framework for generative AI-enabled financial workflows. The framework translates core model risk management principles into a layered control structure for generative AI applications that operate outside the formal model boundary but remain embedded within regulated banking processes. GAICF provides a practical approach for financial institutions seeking to align emerging generative AI governance practices with the risk-based supervisory expectations reflected in SR 26-2. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.04103 |
| By: | Colliard, Jean-Edouard; Zhao, Junli |
| Abstract: | This paper studies how artificial intelligence (AI) affects the finance labor market when humans and AI perform different tasks in investment projects, and workers earn agency rents that grow with project size. We identify two key effects of AI improvement: A freeriding effect raises worker rents by increasing the probability of successful investment when the worker shirks; A capital reallocation effect shifts investment toward workers with higher or lower rents, depending on which tasks AI improves. Contrary to standard predictions, AI can raise both worker rents and labor demand. We derive implications for capital allocation, labor demand, compensation, and welfare. |
| Keywords: | Artificial intelligence; Labor markets; Automation |
| JEL: | D21 G20 O33 |
| Date: | 2025–07 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20460 |
| By: | Adrian, Tobias; Mosk, Benjamin; Wu, Jason |
| Abstract: | Generative Artificial Intelligence is starting to more significantly impact capital market trading activities, on the heels of the broad deployment of machine learning that has already revolutionized trading over the past two decades. The impact of Gen AI is characterized by its ability to process vast amounts of data, allowing market participants to translate new information quickly into price signals. While many current changes appear more evolutionary, more revolutionary shifts may emerge, potentially catching policy makers off guard. Gen AI can amplify existing risks but can also introduce entirely new risks. Rapid adoption could challenge whether existing surveillance and regulatory frameworks are sufficient. However, AI could also help regulators keep pace with growing market complexity, though skill gaps may hinder SupTech adoption. |
| Keywords: | Capital markets |
| JEL: | G11 G12 G14 |
| Date: | 2025–10 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20748 |
| By: | Suela Vasil (Departament of informatics, Faculty of Natural Sciences, University of Tirana); Armela Maxhelaku (Department of Civil Law, Faculty of Law, University of Tirana) |
| Abstract: | Since late 2022, the rapid evolution of generative artificial intelligence and large language models has significantly accelerated the integration of AI into FinTech services, including credit scoring, fraud detection, algorithmic trading, and regulatory compliance. This rapid expansion of this literature identifies the need for taxonomic mapping of AI methods to FinTech application domains. In this article we have applied PRISMA 2020 guideline to peer-reviewed articles indexed in Scopus-and published between 2024 and early 2026? Using a systematic search strategy? 388 records were identified through database searching? Out of these articles? 144 articles met the eligibility criteria and were included in the review? Data were collected using a structured a coding sheet and synthesized through the taxonomic cross-tabulation of AI categories and FinTech application domains? The results show that machine learning? deep learning and natural language processing are the most frequently applied AI models and Random Forest? Long Short-Term Memory (LSTM) and BERT are the most applied AI algorithms in FinTech applications? The FinTech domains that are most heavily deployed are credit scoring and lending? fraud detection and security and cryptocurrency and blockchain applications? This article provides an AI-FinTech taxonomy that could serve as an evidence-based reference for academics? practitioners? and policymakers for the adoption of artificial intelligence in financial services? |
| Keywords: | Artificial intelligence, FinTech, Machine learning, Deep learning, Natural language processing |
| JEL: | C45 G20 O33 |
| URL: | https://d.repec.org/n?u=RePEc:sek:iefpro:15817211 |
| By: | Bartosz Zi\'o{\l}ko; Kacper Dobrzeniewski |
| Abstract: | In this study, we examine the opportunities brought by Large Language Models (LLMs) to various aspects of fundamental analysis of companies based on their reports as well as data and documents describing macroeconomic situation like GDP and inflation changes as well as documents filled to the U.S. Securities and Exchange Commission (SEC) which can be found in EDGAR. We were preprocessing those data and than sending via API to gpt-4o model in a Retrieval-Augmented Generation (RAG) like regime. We prepared as well a document describing an exemplar investor knowledge based on Kitchin cycles. We were scanning data important for analysis of 9 companies for 4 weeks. Using LLM we were producing automatic briefs about them. They were sent to nine participants who are individual investors to evaluate usefulness of such approach to data analysis. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.09121 |
| By: | Hoyoung Lee; Suhwan Park; Seunghan Lee; Jun Seo; Jaehoon Lee; Sungdong Yoo; Minjae Kim; CheolWon Na; Zhangyang Wang; Zach Golkhou; Minkyu Kim; Sotirios Sabanis; Alejandro Lopez-Lira; Dhagash Mehta; Soonyoung Lee; Chanyeol Choi; Wonbin Ahn; Yongjae Lee |
| Abstract: | Financial decision-makers face more information than they can directly inspect, making context compression necessary. Yet when large language models (LLMs) compress financial source material, they can alter the investment judgment supported by the original source. We frame this problem as information fidelity: compression loses fidelity when it changes the decision induced by the source. In agentic systems, such losses may recur across intermediate steps and amplify throughout the decision process. Across financial filings and earnings-call transcripts, we find that LLM-based compression can produce fluent and factually plausible compressed contexts that nevertheless alter downstream decisions. We analyze two diagnostic patterns associated with fidelity loss: decontextualization, where salient evidence is retained but separated from the caveats and contextual qualifiers needed for correct interpretation, and model dependency, where different compressors expose different views of the same source. We then propose Agentic Context Compression, which generates multiple candidate compressions and audits their disagreements against the original source. Our results suggest that financial compression should be evaluated not only by efficiency or factuality, but also by its ability to preserve decision-relevant context. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.29251 |
| By: | Aquilina, Matteo; Araujo, Douglas; Gelos, Gaston; Park, Taejin; Perez-Cruz, Fernando |
| Abstract: | Predicting financial market stress has long proven to be a largely elusive goal. Advances in artificial intelligence and machine learning offer new possibilities to tackle this problem, given their ability to handle large datasets and unearth hidden nonlinear patterns. In this paper, we develop a new approach based on a combination of a recurrent neural network (RNN) and a large language model. Focusing on deviations from triangular arbitrage parity (TAP) in the Euro-Yen currency pair, our RNN produces interpretable daily forecasts of market dysfunction 60 business days ahead. To address the “black box†limitations of RNNs, our model assigns data-driven, time-varying weights to the input variables, making its decision process transparent. These weights serve a dual purpose. First, their evolution in and of itself provides early signals of latent changes in market dynamics. Second, when the network forecasts a higher probability of market dysfunction, these variable-specific weights help identify relevant market variables that we use to prompt an LLM to search for relevant information about potential market stress drivers. |
| JEL: | G14 G15 G17 |
| Date: | 2025–10 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20768 |
| By: | Ralph S. J. Koijen; Bradford Levy |
| Abstract: | Evaluating optimized AI systems for asset pricing is fundamentally difficult for two reasons. First, models are trained on all data, implying that any backtest or analysis using historical data suffers from look-ahead bias. In addition, markets are reflexive — as investors adopt AI, prices adjust — which may erode the very patterns the AI system was trained to exploit. We introduce a real-time, out-of-sample benchmark designed to sidestep both problems. The benchmark measures how well AI systems can explain contemporaneous stock returns around earnings announcements using only information available at announcement time, including the text of the announcement itself. Applying this benchmark to a range of agentic AI systems — which extract structured signals from earnings call transcripts and optimize over those signals — we find that the best-optimized systems more than double the explained variation in returns relative to standard benchmarks (R2 increasing from 8% to close to 20%). We show that AI-based optimization can deliver efficiency gains relative to traditional machine learning methods while also improving interpretability as our approach produces human-readable economic mechanisms that explain price movements. These learned rules can be compared to the drivers of realized returns in existing asset pricing models to identify missing sources of variation in a data-driven, self-evolving way that integrates empirical learning with economic structure. We release an SDK for researchers to improve on our results. Saturating this benchmark would represent fundamental progress in understanding how capital markets process firm-level information. |
| JEL: | C10 G1 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35431 |